Support for Sin Taxes

One control and four information treatments.

2021-12-09

2022-01-31

Political preferences:
(1) Support to introducing a federal tax on sugary beverages in the United States, on a 5-point Likert scale.
(2) Preferred magnitude of sugary taxes in US cents.
(3) Revealed willingness to pay for a donation to a NGO.
(4) Support to introducing a sugary drink tax in another federal state.

Arguments/Channels: For each of the arguments, we include two Likert-scale questions asking whether individuals agree or disagree. We form a standardized index that sums up their approval.

As secondary analyses, we will estimate heterogenous treatment effects by income, gender, party affiliation, self-reported consumption, self-control, and BMI.

Moreover, we will analyze heterogenous treatment effects by a libertarian index and a paternalism index in an exploratory way. These indices are based on a battery of statements measuring political attitudes.

Using the control treatment, we will do exploratory analyses to investigate how sugary drink tax preferences correlate with other variables. For example, we test whether untreated respondents use the economic arguments in the free-text part, and how the approval ratings of the arguments correlate with policy preferences. Moreover, we analyze how their guess of others’ sugary drink consumption and the guessed difference in consumption between income groups correlates with the approval to the arguments and policy preferences. We will also analyze how their political attitudes, measured at the end of the survey, correlates with soft drink tax support.

We plan to use machine learning methods to analyze free-text answers about the importance of the arguments.

Randomized information treatments in a survey experiment

After collecting information about demographics, subjects are given the possibility to write down their opinions about sugary taxes in free-text fields. Subjects are then treated with information about each of the arguments in a cross-subject design (hence, there are four information treatments and one control). After reading the arguments, subjects respond to an incentivized quiz question, addressing subjects’ beliefs about a basic fact related to the argument. For the self-control treatment, respondents are asked to guess the self-reported degree of self-control of participants of the pre-survey. For the misperception treatment, participants have to guess the share of participants in the pre-survey that underestimated the weight implication of sugary drinks. For the externality argument, we ask them about research estimates about health externalities. For the regressivity treatment, we ask about the percentage of income that poor consumers spend on sugary beverages relative to rich consumers.

We then survey subjects’ approval of the economic arguments (channels). Finally, we survey their preferences over the introduction of sugary drink taxes, both unincentivized (self-reported policy approval) and incentivized (donation to an organization that lobbies for the introduction of a federal sugary drinks tax in the US).

For the experimental intervention part, we test the following hypotheses (relative to the control):
H1: Explaining the self-control argument (a) increases approval to the self-control arguments and (b) increases support for sugary drink taxes.
H2: Explaining the misperception argument (a) increases approval to the misperception arguments, and (b) increases the support for sugary drink taxes.
H3: Explaining the health cost externalities of sugar consumption (a) increases approval to the externality argument, and (b) increases support for sugary drink taxes.
H4: Explaining the regressivity of sugary drink taxes (a) increases support for the regressivity arguments, and (b) decreases support for sugary drink taxes.

Within-treatment, we test whether the belief intensity (how large self-control problems, underestimation of health costs, externalities, and regressivity are estimated to be) is correlated with support for sugary drink taxes.

Randomization will be done using Qualtrics using the option “Evenly present elements.”

Individual

No

We aim to collect 4,400 observations in total

We aim to collect 4,400 observations in total

1,200 in the control, and 800 per information treatment

Based on our pilot data, we calculate a MDE of 0.14 of a standard deviation for the attitudes/channels. For the preferred tax magnitude, we calculate a MDE of 4.39 ct, and for the donation, we calculate a MDE of 3.3 ct.

PAP_Support_Sin_Taxes.pdf